[Mlir-commits] [mlir] [mlir][ExecutionEngine] Add LevelZeroRuntimeWrapper. (PR #151038)

[Md Abdullah Shahneous Bari]

================
@@ -0,0 +1,56 @@
+// RUN: mlir-opt %s -pass-pipeline='builtin.module(spirv-attach-target{ver=v1.0 caps=Addresses,Int64,Kernel},convert-gpu-to-spirv{use-64bit-index=true},gpu.module(spirv.module(spirv-lower-abi-attrs,spirv-update-vce)),func.func(llvm-request-c-wrappers),convert-scf-to-cf,convert-cf-to-llvm,convert-arith-to-llvm,convert-math-to-llvm,convert-func-to-llvm,gpu-to-llvm{use-bare-pointers-for-kernels=true},gpu-module-to-binary,expand-strided-metadata,lower-affine,finalize-memref-to-llvm,reconcile-unrealized-casts)' \
+// RUN: | mlir-runner \
+// RUN:   --shared-libs=%mlir_levelzero_runtime \
+// RUN:   --shared-libs=%mlir_runner_utils \
+// RUN:   --entry-point-result=void \
+// RUN: | FileCheck %s
+
+module @add attributes {gpu.container_module} {
+  memref.global "private" constant @__constant_2x2x2xf32_0 : memref<2x2x2xf32> = dense<[[[1.1, 2.2], [3.3, 4.4]], [[5.5, 6.6], [7.7, 8.8 ]]]>
+  memref.global "private" constant @__constant_2x2x2xf32 : memref<2x2x2xf32> = dense<[[[1.2, 2.3], [4.5, 5.8]], [[7.2, 8.3], [10.5, 11.8]]]>
+  func.func @main() {
+    %0 = memref.get_global @__constant_2x2x2xf32 : memref<2x2x2xf32>
+    %1 = memref.get_global @__constant_2x2x2xf32_0 : memref<2x2x2xf32>
+    %2 = call @test(%0, %1) : (memref<2x2x2xf32>, memref<2x2x2xf32>) -> memref<2x2x2xf32>
+    %cast = memref.cast %2 : memref<2x2x2xf32> to memref<*xf32>
+    call @printMemrefF32(%cast) : (memref<*xf32>) -> ()
+    return
+  }
+  func.func private @printMemrefF32(memref<*xf32>)
+  func.func @test(%arg0: memref<2x2x2xf32>, %arg1: memref<2x2x2xf32>) -> memref<2x2x2xf32> {
+    %c2 = arith.constant 2 : index
+    %c1 = arith.constant 1 : index
+    %mem = gpu.alloc host_shared () : memref<2x2x2xf32>
+    memref.copy %arg1, %mem : memref<2x2x2xf32> to memref<2x2x2xf32>
+    %memref_0 = gpu.alloc host_shared () : memref<2x2x2xf32>
+    memref.copy %arg0, %memref_0 : memref<2x2x2xf32> to memref<2x2x2xf32>
+    %memref_2 = gpu.alloc host_shared () : memref<2x2x2xf32>
+    %2 = gpu.wait async
----------------
mshahneo wrote:

Yes, for the gpu.wait op to return a async token, async must be added.

https://github.com/llvm/llvm-project/pull/151038